Multi-dimensional quality of service driven robust resource allocation method for space-air-ground integrated fso / rf

Abstract

The present application relates to a kind of multi-dimensional quality of service driven FSO / RF space-ground integration robust resource allocation method, belong to wireless communication network technical field, the existing FSO / RF SAGIN cloud edge collaborative computing method cannot consider the efficiency of computing task processing, timeliness and computing task transmission reliability and the problem of needing to rely on accurate CSI condition to carry out computing task processing is solved.This application first establishes FSO / RF SAGIN cloud edge collaborative computing network model, and constructs optimization problem OP1 based on the model, to realize the maximum long-time computing task processing amount of all users as the goal, solves optimization problem OP1 using DDPG deep learning method, and the optimal solution is obtained, i.e.FSO / RF space-ground integration robust resource allocation scheme.The present application can ensure the realization of high quality of service FSO / RF space-ground integration cloud edge collaborative computing from multiple dimensions under uncertain channel conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication networks, in particular to a multi-dimensional quality of service driven FSO / RF space-air-ground integrated robust resource allocation method. BACKGROUND

[0002] Future 6G networks will not only provide seamless connectivity worldwide, but also integrate massive computing resources to support ubiquitous device interconnection and real-time data processing, thereby promoting the development of emerging fields such as smart cities, remote medical care, and intelligent agriculture. Space-air-ground integrated networks (SAGIN) empowered by cloud-edge collaborative computing play an important role in 6G networks as they can meet these demands. However, current SAGIN mainly uses radio frequency (RF) transmission links, which face the problems of spectrum resource scarcity and difficulty in secure transmission of computing tasks between satellites and unmanned aerial vehicles. Therefore, applying free space optical (FSO) communication technology with ultra-wideband spectrum resources and high security to SAGIN cloud-edge collaborative computing, i.e., FSO / RF SAGIN cloud-edge collaborative computing technology, is an effective solution.

[0003] Achieving efficient, timely computing task processing and reliable computing task transmission is a basic requirement for FSO / RF SAGIN cloud-edge collaborative computing. However, existing technologies only focus on one aspect, such as improving computing task processing efficiency while ignoring task transmission reliability and processing timeliness, or only considering reliable computing task transmission while ignoring task processing efficiency and timeliness. In addition, the realization of reliable, efficient, and timely FSO / RF SAGIN cloud-edge collaborative computing requires channel state information (CSI) of the transmission channel, and existing FSO / RF SAGIN cloud-edge collaborative computing methods are analyzed under the assumption of known accurate CSI. However, due to the influence of channel estimation error and quantization error, the channel has uncertainty, and it is difficult to obtain accurate CSI information in actual application. Therefore, existing technologies cannot simultaneously achieve reliable, efficient, and timely FSO / RF SAGIN computing task processing under uncertain CSI conditions. SUMMARY

[0004] To solve the problems that the existing FSO / RF SAGIN cloud edge collaborative computing method cannot consider the efficiency, timeliness of computing task processing and reliability of computing task transmission and needs to rely on accurate CSI conditions for computing task processing, the present application provides a multi-dimensional quality of service driven FSO / RF space-air-ground integrated robust resource allocation method, which can guarantee the realization of high-quality FSO / RF space-air-ground integrated cloud edge collaborative computing from multiple dimensions (reliable, efficient and timely) under uncertain channel conditions.

[0005] To solve the above technical problems, the present application adopts the following technical solution:

[0006] A multi-dimensional quality of service driven FSO / RF space-air-ground integrated robust resource allocation method, which comprises the following steps:

[0007] Step 1: Establish a FSO / RF SAGIN cloud edge collaborative computing network model, which includes a satellite, a UAV, a cloud server and K ground Internet of Things users. The satellite is used to provide comprehensive services covering the region of interest, and the UAV serving as an edge server is equipped with a cache device and a computing device, which are used to provide edge computing and task caching for the ground Internet of Things users. Each ground Internet of Things user generates a delay-sensitive computing task, and offloads the computing task not placed in the user's computing cache queue to the UAV through an RF link. The UAV forwards the computing task not placed in the UAV's computing cache queue to the cloud server through the satellite using an FSO link, wherein the FSO link and the RF link both use an adaptive modulation and coding method.

[0008] Step 2: Based on the model established in step 1, an optimization problem OP1 is constructed, which is to maximize the total amount of long-time computing task processing of all users by optimizing the allocation of bandwidth of the UAV to each ground Internet of Things user, the offloading location of the computing task, the flight trajectory of the UAV and the modulation and coding format under the constraints of computing task transmission error frame rate and processing delay and under the condition of uncertain CSI information.

[0009] Step 3: The DDPG deep learning method is used to solve the optimization problem OP1 to obtain an optimal solution, which is the FSO / RF space-air-ground integrated robust resource allocation scheme.

[0010] The method provided by the present application can realize high-quality FSO / RF SAGIN cloud edge collaborative computing from multiple dimensions, and has the following advantages:

[0011] (1) The present application can simultaneously realize reliable computing task transmission, computing tasks can be executed within the maximum delay allowed, and the execution efficiency of computing tasks is high.

[0012] (2) The application considers the uncertainty of CSI information when optimizing network parameters, so it can adjust network parameters under uncertain CSI information conditions, which meets the requirements of actual communication environment;

[0013] (3) Since the FSO communication mode with large capacity and secret transmission advantage is adopted for the ultra-long distance task transmission between the satellite and the unmanned aerial vehicle, the application can realize large-scale computing task transmission, and can keep the task information that does not want to be leaked secret. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A flowchart of the multi-dimensional quality of service driven FSO / RF space-air-ground integrated robust resource allocation method according to the application;

[0015] Figure 2 A schematic diagram of the FSO / RF SAGIN cloud edge collaborative computing network model;

[0016] Figure 3 An Actor-Critic network architecture diagram for computing task offloading, bandwidth allocation and unmanned aerial vehicle flight trajectory optimization based on DDPG. DETAILED DESCRIPTION

[0017] The technical solutions of the application will be described in detail below with reference to the preferred embodiments and the accompanying drawings.

[0018] Referring to Figure 1 , the embodiment provides a multi-dimensional quality of service driven FSO / RF space-air-ground integrated robust resource allocation method, which specifically includes the following steps 1 to 3.

[0019] Step 1: Establishing an FSO / RF SAGIN cloud edge collaborative computing network model, as shown in Figure 2 The model includes: a satellite, which can provide comprehensive services covering the region of interest, and connects the Internet of Things devices and the cloud server through the satellite backbone network; an unmanned aerial vehicle equipped with cache devices and computing devices, which serves as an edge server and provides edge computing and task caching for ground Internet of Things users (IOT users), and can provide f e operations per second; a cloud server, which can provide f c operations per second; K ground Internet of Things users with computing capabilities, and the set of ground Internet of Things users is represented as Each user can provide f u operations per second.

[0020] Each terrestrial IoT user generates latency-sensitive computational tasks and makes real-time decisions about whether to process them locally. If the terrestrial IoT user can complete the computational task, it is placed in the user's computational cache queue, awaiting execution. Otherwise, it is sent to the user's transmission cache queue, awaiting offload to the drone edge server. In this embodiment, the terrestrial IoT user uses an RF link to offload computational tasks from its transmission cache queue to a drone. The drone then collects computational tasks from K terrestrial IoT users and decides whether to execute them locally or offload them to a cloud server. If the task is to be executed locally, it is placed in the drone's computational cache queue, awaiting execution. Otherwise, it is sent to the drone's transmission cache queue, awaiting offload. In this embodiment, the drone uses an FSO link to forward computational tasks from its transmission cache queue to the cloud server via satellite. This embodiment considers that the FSO link between the satellite and the cloud server provides sufficient bandwidth resources to promptly forward computational tasks to the cloud server, so a cache queue is not set up in the satellite. Furthermore, this embodiment also considers that the cloud server has a sufficient number of computing devices, so a cache queue is not set up in the cloud server.

[0021] This embodiment adaptively selects a modulation mode and channel coding rate that can achieve reliable and efficient task transmission based on the estimated CSI of the transmission link to perform task transmission. RF -QAM, channel coding rate is r RF , in The selection range of RF link modulation and coding parameters is given. The FSO link between UAV and satellite, satellite and cloud server adopts adaptive quadrature amplitude modulation m FSO -QAM, The selection range of FSO link modulation and coding parameters is given.

[0022] The present invention adopts the time slot analysis model to divide the total optimization time period into T time slots, each time slot lasts for τ, and its set is expressed as The computational task generated by terrestrial IoT user k in each time slot is Among them, D k (t) is the number of bits of computing task data generated by the user, which is subject to the parameter λ k Poisson distribution; C k (t) is the number of operations required to perform each bit of data; T k (t) is the maximum delay allowed for executing the task. The calculation buffer queue length of each time slot user for The length of the user's transmission buffer queue For where η(t) is an indicator variable that indicates whether the computing task is executed within user k, η(t) is 1 to indicate execution at the user, denotes the negation of η(t), that is, offloading to the UAV to execute; β is the number of operations required to execute 1 bit of data; is the RF link task transmission rate between user k and the UAV; τ is the time of one time slot; denotes the floor symbol. The length of the computing cache queue of the UAV in each time slot is l k (t)∈{0,1} is an indicator variable that indicates whether the computing task is offloaded to the UAV edge server for execution, l k (t) is 1 to indicate execution at the cloud server. The length of the transmission cache queue of the UAV is:

[0023]

[0024] is an indicator variable that indicates whether the computing task is offloaded to the cloud server for execution, is 1 to indicate execution at the cloud server; is the FSO link transmission rate between the UAV and the satellite; is the FSO link transmission rate between the satellite and the cloud server.

[0025] Step 2: According to the FSO / RF SAGIN cloud edge collaborative computing network model constructed in step 1, in order to simultaneously achieve reliable task transmission and efficient and timely computing task processing under uncertain CSI conditions, the present application simultaneously optimizes the allocation of bandwidth of the UAV to each user, the offloading location of the computing task (user, edge server, cloud server), the flight trajectory of the UAV and the modulation and coding format under the constraints of computing task transmission error frame rate and processing delay, and under the condition of uncertain CSI information, so as to achieve the goal of maximizing the long-time computing task processing amount. The optimization problem constructed in this step is:

[0026]

[0027] In this problem optimization goal, is the execution computing task amount of the cloud server at time t, R f (t) is the transmission rate, different links have different transmission rates, is the execution computing task amount of the edge server at time t, is the execution computing task amount of user k at time t, is the modulation order and channel coding rate adopted by the i-th user, and are the modulation orders of the FSO links between the UAV and the satellite, and between the satellite and the cloud server, respectively. In the network model, the sender of the computing task selects the appropriate modulation and coding format and channel coding rate according to the uncertain CSI information. In addition, the application considers that as long as the user can process the newly generated computing task in time, the task is executed at the user (determined by whether the delay in the computing cache queue exceeds the maximum delay of the task), and for this purpose, the decision of whether to execute the computing task locally is not optimized in the above optimization problem OP1.

[0028] In the constraint conditions of the optimization problem, C1, C2 and C3 are the computing task transmission reliability constraints, respectively requiring that the maximum computing task transmission frame error rate of the RF link and the FSO links between the UAV and the satellite and between the satellite and the cloud server be less than the threshold FER T . is the task transmission frame error rate between the user k and the UAV, is the task transmission frame error rate between the UAV and the satellite, is the task transmission frame error rate between the satellite and the cloud server, and their estimated uncertain ranges of CSI are γ uncR , and and are the signal-to-noise ratios of the links between the user k and the UAV, between the UAV and the satellite, and between the satellite and the cloud server, respectively. C4 requires that the computing task processing delay t k of the user k not exceed the allowed maximum processing delay T k ; C5 requires that the amount of computing tasks offloaded by the user k to the UAV edge server not exceed the amount of tasks that can be transmitted by the RF link C6 requires that the total bandwidth allocated to the user not exceed the maximum bandwidth B that the UAV can provide, where B k (t) is the bandwidth of the user k; C7 and C8 are constraints on the flight angle θ u (t) and the speed v u (t) of the UAV, where v max is the maximum flight speed of the UAV; C9 constrains the flight area Q u of the UAV, q u (t) is the position coordinate of the UAV; C10 and C11 constrain the selection range of the adaptive modulation and coding parameters; C12-C14 constrain the computing offloading strategy, requiring that the task can only be offloaded to one of the cloud server or the edge server, l(t) = {l1(t),...,l k (t),...,l K(t)} is the edge server offloading indication vector, is the cloud server offloading indication vector.

[0029] The expression of the frame error rate is the premise of solving the optimization problem, and the expressions of each parameter will be given below.

[0030] The frame error rate of the RF link between user k and the UAV for task transmission is:

[0031]

[0032] where a, b and c are fitting parameters, I M () is the average mutual information of the modulation and coding scheme, is the average mutual information when using m RF -QAM modulation method, and its calculation method can be referred to in “IEEE Journal on selected areas in communications, 2008, 26(8): 1599-1606.”. is the signal-to-noise ratio of the RF transmission link between user k and the UAV, P us is the transmit power of the user, is the receive noise power of the UAV, is the RF channel attenuation, which is jointly determined by the fixed attenuation and the random attenuation The random attenuation obeys the Nakagami-m distribution, and the fixed attenuation can be expressed as:

[0033]

[0034] where u k (t) is the position vector of user k, H is the height of the UAV, and ρ0 is the channel attenuation of 1 m transmission.

[0035] The frame error rate FER of the FSO link for task transmission f is:

[0036]

[0037] In calculating the frame error rate of the FSO link between the UAV and the satellite , γ f in formula (4) takes the value of In calculating the frame error rate of the FSO link between the satellite and the cloud server , γ f in formula (4) takes the value of where a M , gM and γ pM are fitting parameters, M is the FSO link modulation order, is the FSO link signal-to-noise ratio. P is the transmit power, is the receive noise. is the FSO link attenuation, which is jointly determined by a fixed attenuation and a random attenuation The random attenuation obeys a Log-normal distribution, and the fixed attenuation is determined by the transmit antenna gain G Tx , the receive antenna gain G Re , the free space attenuation the atmospheric attenuation Att Atm and the margin of transmission M A .

[0038] The computation task transmission rate of the RF link between user k and the UAV is R sR is the RF link task data transmission baud rate, and is the QAM modulation order and channel coding rate adopted. The computation task transmission rate of the FSO link between the UAV and the satellite is R sFu is the FSO link task data transmission baud rate between the UAV and the satellite. The computation task transmission rate of the FSO link between the satellite and the cloud server is R sFs is the FSO link task data transmission baud rate between the UAV and the satellite.

[0039] The computation task processing delay of user k is:

[0040]

[0041] wherein, is the computation waiting delay in the edge server, is the transmission waiting delay in user k, is the transmission waiting delay in the UAV.

[0042] Step 3: The optimization problem OP1 can be converted into the following four sub-optimization problems:

[0043]

[0044]

[0045] The essence of OP1_1~OP1_3 is to maximize the computation task transmission rate of the FSO link between the satellite and the cloud server under the condition that the maximum frame error rate does not exceed a certain threshold FER Tunder the condition of maximizing the transmission rate. Their solutions are actually the signal-to-noise ratio range corresponding to each transmission format, that is, the switching threshold of the transmission format. In order to solve the switching threshold of the transmission format in OP1_1 is taken as an example for detailed description, where R is the number of elements in the set Since the frame error rate is inversely proportional to the signal-to-noise ratio, the maximum frame error rate corresponds to the minimum signal-to-noise ratio in γ uncR Therefore, by setting The expression of can be derived, and the expressions of the switching thresholds of the transmission formats in OP1_2 and OP1_3 (and and ) can also be derived.

[0046] Substituting the solutions of the optimization problems OP1_1-OP1_3 into OP1_4, OP1_4 can be converted into:

[0047]

[0048] C4-C9, C12-C14

[0049] wherein, is the calculated task transmission rate of the RF link between user k and the UAV when the adaptive transmission is adopted; is the processing time delay of the calculation task of user k when the adaptive transmission is adopted by the system.

[0050] Step 4: The optimization problem OP2 is solved by using the deep learning method of DDPG (Deep Deterministic Policy Gradient). This first needs to convert OP2 into a Markov process, which needs to construct a state space, an action space and a reward function.

[0051] The state space includes the position vector q u (t) of the UAV, the estimated CSI vector The CSI of the FSO link between the UAV and the satellite is The CSI of the FSO link between the satellite and the cloud server is The user transmission buffer queue length The user calculation buffer queue length The UAV transmission buffer queue length The UAV calculation buffer queue length The average remaining time of the calculation task in the user calculation buffer queue wherein, represents the average remaining time of the user in the calculation buffer queue of the i-th user, and the average remaining time of the calculation task in the user transmission buffer queue Among them Represents the average remaining time of tasks in the transmission cache queue of the i-th user, and the average remaining time of tasks calculated by the transmission cache queue of the drone The average remaining time of the calculation task in the UAV calculation cache queue The state space constructed based on these variables is

[0052] The action space includes the bandwidth B(t) allocated by the drone to each user = {B1(t), B2(t), ..., B K (t)}, the unloading location P(t) of the computing task received from each user = {P1(t),…,P i (t),…,P K (t)}, UAV rotation angle θ u (t) and flight speed v u (t), where P i (t) is determined by the cloud server and edge server task offloading indicator variables. The action space constructed based on these variables is

[0053] The reward function is constructed as D p (t) is the task amount of the pth computing task whose processing delay has exceeded its maximum allowable delay at time t; P means that at time t, there are P computing tasks whose processing delay exceeds their maximum allowable delay.

[0054] Based on the state space, action space, and reward function constructed above, the optimization problem OP2 can be transformed into the DDPG-based optimization problem OP3, specifically:

[0055]

[0056] in, α is the discount factor, r t is the reward function, E π [·] indicates statistical averaging operation.

[0057] When using DDPG to solve the optimization problem OP3, the drone can act as an intelligent agent to determine bandwidth allocation, computing task offloading, and drone flight trajectory strategy. This method uses the Actor-Critic network architecture, which consists of an Actor network and a Critic network. The network architecture is as follows: Figure 3 In each time slot, the Actor network obtains the state space S from the communication environment t , and based on the action strategy function μ(S t |θ μ ) Output action where θμ is the Actor prediction neural network parameter. In order to enhance the ability to explore the optimal strategy, the agent superimposes Gaussian white noise N on the output action. t , therefore, the action output by the Actor network is:

[0058]

[0059] Execute the action After that, the agent can obtain immediate reward r from the communication environment t And transition to the next state S t+1 Then, the agent forms the matrix Store in experience sample pool middle.

[0060] Since the action strategy function μ(S t |θ μ ) and θ μ For this reason, it is necessary to set the parameter θ μ Optimize to obtain the action strategy that maximizes the reward value, that is, to obtain the optimal action strategy:

[0061]

[0062] in, is a statistical average operation. The present invention uses the gradient descent method to solve the optimization problem shown in formula (10), and the solution can be expressed as:

[0063]

[0064] Among them, λ∈(0,1) is the learning rate of the Actor network, yes About θ μ The gradient value is specifically expressed as:

[0065]

[0066] Among them, Q(S t ,μ(S t |θ μ )|θ Q ) is the state-action value output by the Critic prediction neural network, θ Q are the parameters of the network.

[0067] θ Q The choice will affect Q(S t ,μ(S t |θ μ )|θ Q) predicts the accuracy of the value, so it is necessary to Q The optimization criterion is to make Q(S t ,μ(S t |θ μ )|θ Q ) prediction error Minimum, that is:

[0068]

[0069] It can be expressed as:

[0070]

[0071] Among them, μ'() and Q'() are the output values ​​of the target neural network of Actor and Critic respectively. The parameters of these two neural networks are θ μ' and θ Q' .

[0072] The optimization problem (13) can also be solved by the gradient descent method, and the result can be expressed as:

[0073]

[0074] υ∈(0,1) is the critic network learning rate, yes About θ Q The gradient value is specifically expressed as:

[0075]

[0076] θ μ and θ Q The optimization process requires θ μ' and θ Q' , the values ​​of these two parameters can be calculated by the following iterative process:

[0077] θ Q' ←ξθ Q +(1-ξ)θ Q' (17)

[0078] θ μ' ←ξθ μ +(1-ξ)θ μ' (18)

[0079] Where ξ is the update rate of the target neural network.

[0080] and The calculation of needs to perform statistical averaging operation. randomly sample experience samples in the experience sample pool, estimate the statistical average value in the experience sample pool by averaging the sampled samples and The statistical average value in the experience sample pool. For this purpose, the agent is equipped with an experience sample pool with a capacity of C This experience sample pool is used to store experience data At each training time slot, the agent draws M groups of experience samples from the experience sample pool , and Based on the M groups of sampled experience samples (mini-batch samples), the and can be expressed as:

[0081]

[0082] The multi-dimensional quality of service driven FSO / RF space-air-ground integrated robust resource allocation method provided by the present application is a multi-dimensional quality of service driven FSO / RF SAGIN robust computing task offloading, bandwidth allocation and unmanned aerial vehicle flight trajectory optimization method. The method can realize high service quality FSO / RF SAGIN cloud edge collaborative computing from multiple dimensions, and has the following beneficial effects:

[0083] (1) The present application can realize reliable computing task transmission, and the computing task can be executed within the maximum delay allowed, and the execution efficiency of the computing task is high;

[0084] (2) The present application considers the uncertainty of CSI information when optimizing network parameters, so it can adjust the network parameters under uncertain CSI information, which meets the requirements of the actual communication environment;

[0085] (3) Since the FSO communication mode with large capacity and secure transmission advantage is adopted for super-long distance task transmission between satellites and unmanned aerial vehicles, the present application can realize large-scale computing task transmission, and can keep the task information that does not want to be leaked secret.

[0086] The technical features of the above-described embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described, but as long as the combinations of the technical features do not contradict, they should be considered within the scope of the present disclosure.

[0087] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A multi-dimensional quality of service driven FSO / RF air-ground integrated robust resource allocation method, characterized by: The following steps are involved: Step 1: Establish an FSO / RF SAGIN cloud-edge collaborative computing network model, which includes satellites, drones, cloud servers, and K terrestrial IoT users. The satellites provide comprehensive services covering the area of ​​interest. The drones, acting as edge servers, are equipped with cache and computing devices to provide edge computing and task caching for terrestrial IoT users. Each terrestrial IoT user generates latency-sensitive computing tasks and offloads these tasks, which are not queued in the user's computational cache, to the drones via RF links. The drones then forward these tasks to the cloud servers via FSO links. Both FSO and RF links employ adaptive modulation and coding. Step 2: Based on the model established in Step 1, an optimization problem OP1 is constructed. This optimization problem OP1 is: Under the constraints of the transmission frame error rate and processing delay of the computing task, and under the condition of uncertain CSI information, the bandwidth allocated by the UAV to each ground IoT user, the offloading location of the computing task, the UAV flight trajectory, and the modulation and coding format are optimized to achieve the goal of maximizing the total long-term computing task processing capacity of all users; Step 3: Use the DDPG deep learning method to solve the optimization problem OP1 and obtain the optimal solution, which is the FSO / RF air-ground integrated robust resource allocation solution.

2. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to claim 1, characterized in that: The expression of the optimization problem OP1 is: In the optimization objective of the optimization problem OP1, is the amount of computing tasks performed by the cloud server in time slot t, τ is the time of a time slot, is the transmission cache queue length of the UAV, is the FSO link transmission rate between the UAV and the satellite, is the FSO link transmission rate between the satellite and the cloud server; is the amount of computing tasks performed by the edge server in time slot t, f e is the number of operations that the drone can provide per second, β is the number of operations required for the user to execute 1 bit of data, is the amount of computing tasks performed by user k in time slot t, Calculate the cache queue length for each time slot user, f u The number of operations that can be provided to each user per second, is the modulation order and code rate used by the i-th user, and are the modulation orders of the FSO links between the UAV and the satellite, and between the satellite and the cloud server; B(t)={B1(t),B2(t),...,B K (t)} is the bandwidth allocated to each user by the drone; In the constraints of the optimization problem OP1, C1, C2, and C3 are the reliability constraints for computing task transmission, which require that the maximum computing task transmission frame error rate of the RF link and the FSO link between the drone and the satellite, and the satellite and the cloud server be less than the threshold FER. T ; is the frame error rate of task transmission between user k and the UAV, The frame error rate of the mission transmission between the UAV and the satellite, is the frame error rate of task transmission between satellite and cloud server, and their estimated uncertainty ranges of CSI information are γ uncR 、 and and are the signal-to-noise ratios of the links between user k and the drone, the drone and the satellite, and the satellite and the cloud server respectively; C4 requires the computing task processing delay t for user k. k Cannot exceed the maximum allowed processing delay T k ; C5 requires the amount of computing tasks that user k offloads to the drone edge server The amount of work that can be transmitted by the RF link cannot be exceeded C6 requires the total bandwidth allocated to the user It cannot exceed the maximum bandwidth B that the drone can provide, where B k (t) is the bandwidth of user k; C7 and C8 are the bandwidth of the UAV flight angle θ u (t) and rate v u (t) constraint, where v max is the maximum flight speed of the drone; C9’s flight area Q u To constrain, q u (t) is the position coordinate of the UAV; C10 and C11 constrain the selection range of adaptive modulation and coding parameters; C12 to C14 constrain the computation offloading strategy, requiring that the task can only be offloaded to one of the cloud server or edge server, l(t) = {l1(t),...,l k (t)..., l K(t)} is the edge server offloading indication vector, is the cloud server offload indication vector.

3. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to claim 2, characterized in that: Frame error rate of the RF link between user k and the UAV for: Among them, a, b and c are fitting parameters; Is to use m RF -Average mutual information when using QAM modulation; is the signal-to-noise ratio of the RF transmission link between user k and the UAV; P us is the user's transmit power; is the UAV receiving noise power; is the RF channel attenuation, which consists of a fixed attenuation and random decay Joint decision.

4. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to claim 2, characterized in that: Frame error rate of mission transmission between drone and satellite for: Frame error rate of mission transmission between satellite and cloud server for: Among them, a M 、g M and γ pM is the fitting parameter; M is the FSO link modulation order; and is the FSO link signal-to-noise ratio, which is P is the transmitting power; is the receiving noise; is the FSO link attenuation, which is composed of fixed attenuation and random decay Joint decision.

5. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to any one of claims 2 to 4, characterized in that: In step 3, when using the DDPG deep learning method to solve the optimization problem OP1, the optimization problem OP1 is first transformed into the following four sub-optimization problems: stC1,C10; stC2,C11; stC4~C9,C12~C14; Sub-optimization problems OP1_1 to OP1_3 represent the maximum frame error rate not exceeding the threshold FER T Under the condition of , the transmission modulation and coding format is adaptively adjusted to maximize the transmission rate. The solution results of sub-optimization problems OP1_1 to OP1_3 are substituted into sub-optimization problem OP1_4, and sub-optimization problem OP1_4 is converted into optimization problem OP2. The expression of optimization problem OP2 is: C4~C9, C12~C14; in, is the transmission rate of the computational task of the RF link between user k and the UAV when adaptive transmission is used; It is the processing delay of the computing task of user k when the system adopts the adaptive transmission mode.

6. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to claim 5, characterized in that: When solving the maximum transmission rate corresponding to the sub-optimization problems OP1_1 to OP1_3, the switching threshold of each transmission format is obtained.

7. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to claim 5, characterized in that: When using the DDPG deep learning method to solve the optimization problem OP2, we first construct the state space, action space and reward function, and convert the optimization problem OP2 into a Markov process.

8. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to claim 7, characterized in that: The constructed state space includes: the position vector q of the drone u (t), estimated CSI vector of K RF links FSO link CSI between drone and satellite FSO link CSI between satellite and cloud server User transmission buffer queue length User calculation cache queue length Drone transmission cache queue length Drone calculation cache queue length Average remaining time of calculation tasks in user calculation cache queue Among them represents the average remaining time of the user in the calculation cache queue of the i-th user and the average remaining time of the calculation task in the user transmission cache queue Among them Represents the average remaining time of tasks in the transmission cache queue of the i-th user, and the average remaining time of tasks calculated by the transmission cache queue of the drone The average remaining time of the calculation task in the UAV calculation cache queue The constructed action space includes: the bandwidth B(t) allocated by the drone to each user = {B1(t), B2(t), ..., B K (t)}, the unloading location P(t) of the computing task received from each user = {P1(t),…,P i (t),…,P K (t)}, UAV rotation angle θ u (t) and flight speed v u (t), where P i (t) is determined by the cloud server and edge server task offloading indicator variables; The reward function constructed is Among them, D p (t) is the task amount of the pth computing task whose processing delay has exceeded its maximum allowable delay at time t; P means that at time t, there are P computing tasks whose processing delay exceeds their maximum allowable delay.

9. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to claim 8, characterized in that: Based on the constructed state space, action space, and reward function, the optimization problem OP2 is transformed into the DDPG-based optimization problem OP3. The expression of the optimization problem OP3 is: in, α is the discount factor, E π [·] indicates statistical averaging operation.

10. The multi-dimensional service quality driven FSO / RF air-ground integrated robust resource allocation method according to claim 9, characterized in that: When DDPG is used to solve the optimization problem OP3, an Actor-Critic architecture is adopted, with the drone as the intelligent agent to determine the bandwidth allocation, computing task offloading location, and drone flight trajectory strategy.

Images